In order to study the metabolic transcription of different species of Accumulibacter under low-phosphate conditions, the results from two experiments were compared. The two reactors were operated apart from each other, with a different medium composition and an inoculum sampled from the same wastewater treatment plant, but at different moments.

In experiment RA, the carbon source came from acetate and propionate at equal proportions on a COD basis (COD 300 mg/L), and the nitrification was inhibited by adding allylthiourea in the medium. For 103 days, the COD/P-PO₄ (mg/L COD/mg/L P-PO₄) ratio was maintained at a value of 12. Then, the phosphate concentration in the medium was decreased to reach a COD/P-PO₄ ratio of 200. After the phosphate reduction, a soft conditioning cycle was operated by inverting the settling and withdrawal phases with the aerobic phase for two cycles repeated twice in two weeks. The experiment was previously described in detail in (Cardona et al., 2025).

The experiment RC was operated for several years in the lab before being used for this experiment. Day 0 corresponds to the start of the present study. The carbon source came from acetate and glucose, and the nitrification was not inhibited in that experiment. After 76 days of operation, the COD/P-PO₄ ratio was progressively modified from 20 to 200 for 27 days (day 76 to 103). Then, the initial COD/P-PO₄ ratio of 20 was recovered and maintained for 28 days. A second decrease of the COD/P-PO₄ ratio was initiated on day 131 and maintained for over 42 days. No conditioning cycle was operated in this experiment.

Both experiments were operated in a bubble column sequencing batch reactor (SBR) of 6.2 cm diameter and 2.4 L working volume in fill-draw mode. The temperature was regulated at 18 °C +/−1 °C by recirculating water in the double wall of the reactor. The pH was maintained at 7.5 +/− 0.5 by monitoring and regulating the injection of 1 mM HCl or 1 mM NaOH with an ISFET probe (Endress+Hauser, Switzerland) using a PID process control. Headspace gas recirculation was used to mix the biomass. Nitrogen or air was added to adjust for the concentration, and this was controlled via a PID to maintain an oxygen concentration of 0% under the anaerobic phase or 100% in the aerobic phase. The flow rate of the gas pump was set at 2 L/min. The pO₂ was monitored by an ISFET probe (Endress+Hauser, Switzerland). Data collected by the reactor probes and the control of the different pumps were relayed to/from a computer through relay modules (WAGO, Switzerland) and processed using the software DAQFactory (AzeoTech, Inc.).

A typical cycle corresponded to 5 to 10 min of sparging nitrogen gas for RA and RC, respectively, 60 min of feeding under anaerobic mixing conditions from the bottom of the reactor; 30 min of anaerobic phase; 120 min of aerobic phase by sparging compressed air; 15 min of nitrogen sparging for RA exclusively; 10 min of settling and withdrawal of half of the reactor working volume. The hydraulic retention time was set at 9.5 h, and the solid retention time was set at 21 days by sampling the mixed liquor three times a week at the end of the aerobic phase. The sampling was also adjusted according to the biomass concentration in the reactor to avoid a washout of the biomass. The experiment setup and design are described in the Supplementary Figure 1.

The sludge used as inoculum was collected in the anaerobic tank of the Thunersee wastewater treatment plant (Thun, Switzerland), which performs biological phosphorus removal.

For experiment RA, the reactor influent was created by mixing two 8.89 times concentrated solutions of C and NP and Milli-Q water. Concentrated solution C contained 5.67 g/L C₂H₃O₂Na-3H₂O, 2.28 g/L C₃H₅O₂Na, 0.889 g/L MgSO₄-7H₂O, 2.2 g/L MgCl₂-6H₂O, and 0.4 g/L CaCl₂-H2O. Concentrated solution NP contained 1.671 g/L K₂HPO₄, 0.649 g/L KH₂PO₄, and 0.048 g/L C₄H₈N₂S to inhibit the nitrification and 50 mL of trace elements solution composed of 16.22 g/L C₁₀H₁₄N₂Na₂O₈-H₂O, 0.44 g/L ZnSO₄-7H₂O, 1.012 g/L MnCl₂-4H₂O, 7.049 g/L (NH₄)₂Fe(SO₄)2-6H₂O, 0.328 g/L (NH₄)₆Mo₇O₂₄-4H₂O, 0.315 g/L CuSO₄-5H₂O, and 0.322 g/L CoCl₂-6H₂O. Both solutions were autoclaved in 10 L glass bottles. Before use, 250 mL of bicarbonate solution composed of 0.933 g/L NH₄HCO₃ and 0.533 g/L KHCO₃ was added to the NP solution to reach a final volume of 10 L. At each cycle, 120 mL of concentrated solutions C and NP were mixed with 960 mL of distilled water to feed the reactor and achieve a final chemical oxygen demand (COD) concentration of 300 mgO₂/L in the SBR.

For experiment RC, the synthetic wastewater contained a COD: NH₄-N: PO₄-P ratio of 400:28:20 mg/L. The final medium resulted from mixing two 24-times concentrated solutions with milli-Q water. Concentrated solution A contained the carbon sources with 11.6 g/L acetate (C₂H₃O₂Na-3H₂O), 4.9 g/L glucose (C₆H₁₂O₆-H₂O), 2.1 g/L MgSO₄-7H₂O, 0.3 g/L CaCl₂-2H₂O, 0.8 g/L KCl, 2.5 g/L NH₄Cl, 0.02 g/L yeast extract, and 7 mL trace elements solution. The solution was autoclaved without the glucose that was prepared aside, filtered, and added to the main solution under a laminar flow hood. Concentrated solution B contained the phosphorous source of 2.7 g/L, 0.6 g/L, or 0.2 g/L K₂HPO₄, for a COD/ P-PO₄ ratio of 20, 90, and 200, respectively.

The nutrient removal performance of the reactors was measured on a weekly basis. Samples of 50 mL were taken from the middle of the water column at the end of the anaerobic and aerobic phases, and centrifuged for 5 min at room temperature at 4200 x g. The supernatant was filtered (0.45 μm). A sample of the synthetic reactor influent was also collected and filtered (0.45 μm). The samples were stored at 4 °C until further analysis was conducted. The concentration of the anions (P-PO₄³⁻, N-NO³⁻, and N-NO²⁻) was measured using ionic chromatography (IC, ICS-90, IonPacAS14A column) with an electrical conductivity detector (Dionex, Switzerland). The chemical oxygen demand was measured by spectrophotometry using two different kits: LCK514 (100–2000 mgO₂/L) and LCK314 (15–150 mgO₂/L) (Hach, USA), measured on a spectrophotometer DR 3900 (Hach, USA).

The total and volatile solids were determined in the sludge obtained by centrifuging 100 mL of the mixed liquor reactor sample taken at the end of the aerobic phase. The mass of the dried pellet after 12 h of drying at 105 °C yielded the total solids, and the mass loss after 2 h of calcination at 550 °C resulted in volatile solids. The results of the monitoring are summarized in the Supplementary Table 1.

For the metatranscriptomics analysis, biomass samples were collected just before the phosphate concentration was changed and again when the anaerobic release of phosphate ceased. One sample was taken after 15 min of the anaerobic feeding started, and another 15 min after the aerobic phase started. Samples were collected during three consecutive cycles for each time point, leading to a total of 12 samples per experiment. An aliquot of 15 mL of mixed liquor was sampled, put on ice, and quickly centrifuged for 1 min at 4 °C and 4,200 x g. The pellet was resuspended in 2 volumes (g pellet:ml volume) of RNA protect Tissue (Qiagen, Germany) to protect the RNA from degradation and homogenized by passing 3 times through a needle (26G). After an incubation at room temperature (RT) for 5 min, the sample was centrifuged for 5 min at RT and 5,000 x g, and the supernatant was discarded. The pellet was snap-frozen in liquid nitrogen and stored at −80 °C until RNA extraction was performed. The RNA preservation was tested beforehand to evaluate the degradation and quantity of RNA after a long period of storage, and showed good preservation capacity (data not shown).

The protocol for the RNA extraction was described in Cardona et al. (2025). Briefly, the samples were resuspended in 0.5 mL of TRIzol (#15596–0026, Invitrogen, Fisher Scientific AG, Switzerland) and incubated for 5 min at RT. Then, 0.1 mL of chloroform 99 + % was added, and the mixture was vortexed for 15 s and incubated for 2 min at RT before being centrifuged for 15 min at 15500 x g at 4 °C. The upper portion was recovered and mixed with 400 μL of 100% ethanol. RNA was purified using an RNA purification kit (Direct-zol RNA Miniprep #R2050, Zymo Research, Germany) following the manufacturer’s recommendations, except that centrifugation was performed for 1 min at 13000 x g. DNA was removed using a TURBO DNA-free™ kit (#AM1907, Thermo Fisher Scientific, Switzerland) following the manufacturer’s recommendations. RNA was purified by adding 76 μL magnetic beads (Agencourt RNA Cleaner XP, #A63987, Beckman Coulter) to the extracted RNA. The RNA was washed by alternating three times between the addition of 70% ethanol solution and removal after 10 min on a magnetic rack. After removing the ethanol, 32 μL of RNAse and DNase-free water was added to the pellet out of the rack and resuspended 10 times by up and down. The samples were incubated for 1 min before being returned to the rack for 1 min. Finally, the supernatants were collected. The quality of the extraction and the absence of DNA on the RNA-extracted samples were assessed by carrying out a PCR. The reaction mix was composed of 2 μL of 5X MyFi reaction buffer and 0.4 μL of the following primers: 27F-FTCGTCGGCAGCGTCAGATGTGTATAAGAGACAGAGMGTTYGATYMTGGCCTCAG and 338R-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGCTGCCTCCCGTAGGAGT (Microsynth, Switzerland), 0.4 μL of MyFi DNA polymerase (#BIO25049, Labgene Scientific SA, Switzerland), and 5.8 μL of nuclease-free demineralized water. The PCR program was composed of the following steps: 95 °C for 1 min, 30 cycles of 15 s at 95 °C, 15 s at 56 °C, and 15 s at 75 °C. The PCR products were checked on an agarose gel of 1.5% with a migration at 100 V. Bacterial ribosomal RNA (rRNA) was removed using a QiaSeqFastSelect 5S/16S/23S kit (#335925, Qiagen) following the manufacturer’s recommendations on 1 μg of total RNA (protocol with TruSeq® stranded library preparation) with the following modification: The first step of combined fragmentation and hybridization was performed for 1 min at 89 °C. Libraries were then generated using the TruSeq Stranded mRNA sample preparation kit (#20020594, Illumina, USA) and IDT for Illumina TruSeq RNA UD Indexes (#20022371, Illumina) following the reference guide #1000000040498 for the LS procedure without optional steps. For the clean-up amplified DNA step, the ratio of magnetic beads to PCR products was 0.7, and 20 μL of RSB was added to release the genetic material from the beads. The amplification was quantified with the Qubit dsDNA HS Assay Kit (#Q32854, Life Technologies), and the quality was checked by electrophoresis using the Agilent High Sensitivity DNA Kit (# 5067–4,626, Agilent Technologies). The concentrations of the samples were normalized to 10 nM and pooled. Sequencing analysis was performed at the Lausanne Genomic Technologies Facility, University of Lausanne (Switzerland), on a NovaSeq 6,000 in paired-end mode (2 × 150). A sequencing run was carried out for each experiment separately.

Metagenome-assembled genomes (MAGs) of Accumulibacter and Propionivibrio were obtained from public databases (Martín et al., 2006; Flowers et al., 2013; Skennerton et al., 2015; Welles et al., 2015; Kantor et al., 2015; Albertsen et al., 2016; Parks et al., 2017; Zhang et al., 2017; Arumugam et al., 2019, 2021; Camejo et al., 2019; Ye et al., 2020; Singleton et al., 2021; Lin et al., 2021; McDaniel et al., 2021b, 2021a; Xie et al., 2024a, 2024b) and previous research done by our team (Adler et al., 2022; Saini et al., 2024). The nomenclature and classification in the present study are based on this latest reevaluation (Petriglieri et al., 2022). The quality of the genomes was assessed using CheckM [Parks et al. (2015), v1.2.2] and their average nucleotide identity between them using FastANI [Jain et al. (2018), v1.33]. The genomes were dereplicated based on the results of fastANI, considering 97% as a threshold to separate two genomes. Supplementary Table 2 summarizes the information related to the MAGs, and Supplementary Table 3 contains the fastANI matrix result. A representative genome for each group was selected based on its quality (completeness, contamination, and fragmentation) and the presence of universal marker genes, obtained using fetchMGs (https://github.com/motu-tool/FetchMGs, v1.3) (Supplementary Table 4). DRAM v1.4.6 (Shaffer et al., 2020) and eggNOG-mapper v2.1.11 (Cantalapiedra et al., 2021) were used for gene prediction and annotation. The selected genomes were used as a database for the metatranscriptomics mapping. The annotation file is presented in the Supplementary Table 5.

The quality of the reads was evaluated using FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) at each step of the analysis pipeline. The results were summarized using MultiQC v1.13 (Ewels et al., 2016). Reads were filtered and trimmed using BBDuk from BBMap v39.01 (https://jgi.doe.gov/data-and-tools/software-tools/bbtools/bb-tools-user-guide/bbmap-guide/) using the following parameters: ktrim = r, k = 23, mink = 11, hdist = 1, tpe, tbo for the adapter trimming steps and qtrim = rl, trimq = 20, minlen = 50, maq = 20, maxns = 1 for the quality trimming and filtering. Ribosomal RNA was removed using sortMeRNA [Kopylova et al. (2012), v4.3.6] using the databases for silva-bac-16 s-id90, silva-arc-16 s-id95, silva-euk-18 s-id95, silva-bac-23 s-id98, silva-arc-23 s-id98, silva-euk-28 s-id98, rfam-5 s-database-id98, and fam-5.8 s-database-id98 from Silva (Quast et al., 2012) and rfam (Kalvari et al., 2021). The remaining messenger RNA was mapped onto the metagenomes database of Accumulibacter and Propionivibrio described in section 2.6, using bowtie2 [Langmead and Salzberg (2012), v2.4.1] and the following arguments: very-sensitive mode, X 1000, phred33, and k 35, allowing to search for a maximum of 35 valid alignments for each read, considering a possible multi-mapping in the different MAGs of the database. The alignments were filtered out if the percentage of identity was lower than -p 95 using msamtools (https://github.com/arumugamlab/msamtools, v1.1.0). Supplementary Figure 2 represents the analytical pipeline, and Supplementary Figure 3 and Supplementary Table 6 summarize the number of reads kept at each step. FeatureCount [Liao et al. (2014), v2.0.1] was used to summarize the mapping results into a count table using the following parameters: t CDS, −g ID, -O, −M, -B, −-primary and -Q 2 to account for the uniquely mapping reads only (Supplementary Tables 7, 8 for RA and RC, respectively), discarding the multi-mapping reads from the subsequent analyses.

To ensure that no PAOs or GAOs genomes were missing from the mapping on the selective database, the unmapped reads were blasted using DIAMOND [v2.1.9.163, Buchfink et al. (2021)] in blastx mode (−k 1) to the Swiss-Prot and TrEMBL databases [2024_04, Bairoch (2000)] where the protein sequences with an annotation score of 2, 3, 4, and 5 were selected. The results for experiment RC showed that between 54 to 74% of the reads mapped to the Bacteria domain and 24 to 42% to the Eukaryota domain, whereas really few reads were assigned to viruses and Archaea. From the Bacteria domain, few reads mapped to some GAOs, Contendobacter (around 1%), Competibacteraceae (1.14%), and Propionivibrio (3%). For experiment RA, the unmapped reads were assigned 31% to the Eukaryota domain and 65% to the Bacteria domain. From the Bacteria domain, around 5.6% of the reads mapped to Accumulibacter sp. and 5.6% to Contendobacter but no other PAO or GAOs were identified with a relative abundance higher than 1%. Although some reads mapped to some PAOs and GAOs that were not included in our database, the percentage of mapping was really low, and it was decided to pursue the analysis without adding these genomes.

All the subsequent analyses were conducted using R (v4.3.3) on RStudio (2023.12.1 + 402). A Venn diagram was done using InteractiVenn (Heberle et al., 2015).

The transcription profile of each MAG in the different samples was evaluated by comparing the number of transcribed genes to the sum of counts per MAG and sample. MAGs with a high transcription level in each sample were selected to compare their transcription profile between different conditions or between different MAGs.

On selected MAGs, the low counts were filtered out using filtrExpr from EdgeR with the parameter min.count = 15 [Robinson et al. (2010), v4.0.16]. Then, the zero values were imputed by 0.5, and the counts were normalized by the gene length. Finally, a normalization using universal marker genes was done as described in Salazar et al. (2019), with the difference that the normalization was done at the MAG level. Briefly, each gene count of each MAG was divided by the median count of 10 universal marker genes of the same MAG, previously identified by fetchMGs. The advantage of using the universal marker genes, genes constitutively transcribed in any conditions, compared to more traditional methods, is the estimation of the transcription to the relative number of copies per cell of each MAG, as described by (Sunagawa et al., 2013; Milanese et al., 2019; Salazar et al., 2019). The data obtained was transformed into counts by dividing each value by the maximal value of each MAG and multiplying by 10^{^9}.

Differential gene expression analysis compared the transcription level of one MAG between two conditions (high versus low influent phosphate concentration). In that case, DESeq2 [Love et al. (2014), v1.42.1], edgeR and limma [Ritchie et al. (2015), v3.58.1] were used, and a gene was defined as differentially expressed if it was determined as such in at least two of these methods with an adjusted p-value (Benjamini-Hochberg method) lower than 0.01 and no restriction on the log fold change value.

The transcription profile was also compared between two MAGs, Propionivibrio aalborgensis versus Accumulibacter regalis in the case of experiment RA, and Accumulibacter delftensis (type I) and Accumulibacter phosphatis (type II) in experiment RC. For these comparisons, the counts were summed at the KEGG Orthologs (KO) level per MAG and sample, and edgeR was used to identify the differential gene transcription (adjusted p-value lower than 0.01). For all analyses, comparisons were made for each phase separately (feeding and aerobic). A list of genes of interest is provided in the Supplementary Table 10. The results of the differential gene transcription analyses are provided in Supplementary Tables 11, 12, for RA and RC, respectively.

In two experiments (RA and RC), aerobic granular sludge reactors with biomass enriched in PAOs were switched to an influent low in phosphate. In experiment RA, the carbon sources were acetate and propionate, whereas in experiment RC, acetate and glucose were supplied.

Before changing the influent phosphate (P-IN), the phosphate concentration at the end of the anaerobic phase (P-AN) reached a maximum of 237 and 96.5 mg/L. It decreased during the aerobic phase (P-AE) to as low as 0.80 mg/L and below the limit of detection for RA and RC, respectively (Figure 1).

Once P-IN was reduced to reach a COD/ P-PO₄ ratio of 200, P-AN decreased in both experiments. In RA, from day 150, P-AN decreased to 1 mg/L until day 169, from which P-AN slightly increased to a maximum of 16.6 mg/L. In RC, while P-AN dropped to 1 mg/L in two weeks after the first modification (days 76 to 100), the decrease took less than a week in the second modification (days 130 to the end). The return to the initial P-IN, from days 100 to 130, led to an increase of P-AN of around 100 mg/L, similar to the concentration at the beginning of the experiment.

In both experiments, the carbon removal performance at the end of the anaerobic phase remained higher than 80% irrespective of the P-IN concentration. This result suggests a GAM phenotype, as phosphate release was low, while the carbon was taken up under anaerobic conditions. In order to evaluate the influence of the phosphate concentration on the PAO populations and their metabolism, metatranscriptomics samples were taken before and after P-IN reduction.

To identify the predominant Accumulibacter and Propionivibrio populations and the extent to which their respective transcriptomes were sequenced, the proportion of uniquely mapped reads and the sequencing depth of the metatranscriptomics analysis for each MAG were evaluated. Between the two experiments, a difference in Accumulibacter populations is observable. Before changing the phosphate concentration (Figure 2A, day 103), A. necessarius and A. regalis were the most abundant in RA, with 57 and 16% of the mapped reads, respectively, while A. phosphatis (46%) was the most abundant one in RC (Figure 2B, days 74 and 126). Furthermore, in RA, the reads mapped onto more MAGs compared to RC, where mainly four MAGs captured the reads. After decreasing the influent phosphate concentration, the proportion changed in both experiments. In RA, A. regalis remained abundant (23%), A. necessarius decreased to 4%, while Propionivibrio increased to 58% (Figure 2A, day 182). In RC, A. delftensis became the most abundant species with 84% of the mapped reads (Figure 2B, day 168).

To be able to compare the metabolism of Accumulibacter in the different conditions, a selection of the most transcriptionally active MAGs in all conditions was done by comparing the level of transcription to the number of genes transcribed by the MAG in the different samples (Supplementary Figure 4). In experiment RA, one MAG of Accumulibacter (A. regalis) and one MAG of Propionivibrio (P. aalborgensis) were selected. In experiment RC, two MAGs of Accumulibacter (A. delftensis and A. phosphatis) were kept for further analysis.

The different species of Accumulibacter and Propionivibrio were classified based on the Average Nucleotide Identity (ANI) analysis (Supplementary Figure 5). A. regalis and A. delftensis, highlighted in samples under low-phosphate conditions, are both from type I and belong to clades IA and IC, respectively. On the other hand, A. phosphatis was classified as type IIA.

Table 1 represents the number of genes present and transcribed for each Accumulibacter MAGs. A. delftensis and A. phosphatis have both more than 4,100 genes, more than A. regalis (3746). A total of 88% (A. phosphatis) to 99% (A. delftensis and A. regalis) of these genes were transcribed in any of the experiments, and 80% of the genes annotated with a KO were common between the different MAGs (Table 1 and Supplementary Figure 6). When comparing the transcription profiles of Accumulibacter under low- versus high-phosphate conditions in each experiment, it was observed that few genes were differentially transcribed.

Figure 3 and Supplementary Figures 7, 8 represent the gene transcription of the pathways involved in the EBPR process for both anaerobic and aerobic phases, for the different Accumulibacter species selected in the previous steps. Furthermore, the genes considered as differentially transcribed in high- versus low-phosphate conditions are highlighted. The results obtained for both phases provided similar results, but the level of transcription in the anaerobic and aerobic phases were different which could indicate a difference of transcription between the phases in each condition. However, the analysis was not done to compare the gene transcription between phases, as the main focus of the experiment was the comparison between high- and low-phosphate conditions. Nonetheless, based on previously reported results (Oyserman et al., 2016), a direct comparison was made of gene transcription between phases; some differences should be expected in these experiments, too.

From the pathways involved in the EBPR process, it can be observed that only a few of them were affected by the decrease in the phosphate concentration. Mainly, the phosphate metabolism, a part of the tricarboxylic acid (TCA) cycle, the methylmalonyl (MMC) and ethylmalonyl-CoA (EMC) metabolisms presented a higher number of up-regulated genes in low- versus high-phosphate conditions, while the glycogen branching-debranching, the TCA cycle, the PHA biosynthesis-degradation, the VFA activation, or the glyoxylate shunt were not significantly affected. More specifically, Accumulibacter relied on the high-affinity phosphate transporter under low-phosphate conditions and on the pit system under high-phosphate conditions, as shown by the higher transcription of the pstSABC genes and the pho regulator system. More phosphate-related genes were affected in type I populations compared to type II, which additionally showed lower transcription of pstSABC.

From the pathways potentially involved in balancing the redox under low-phosphate condition, Accumulibacter seemed to rely more on the ethylmalonyl-CoA (EMC) pathway as most of the genes were up-regulated. Conversely, Accumulibacter lacks some genes of the methylmalonyl-CoA (MMC) pathway and did not show significant differences in gene transcription. An exception can be made for Accumulibacter phosphatis type II, which showed a higher transcription of the methylmalonyl-CoA decarboxylase (mmd), responsible for the production of the propionyl-CoA under low-phosphate conditions.

Only A. delftensis (type I) from experiment RC presented differences in the transcription of the genes related to the Embden-Meyerhof-Parnas pathway including phosphoglycerate mutase (pgm), phosphoglycerate kinase (pgk), enolase (eno), pyruvate kinase (pyk) and the VFA transporter (actP).

In experiment RC uniquely, Accumulibacter type I was compared to Accumulibacter type II under high- (day 126) or low-phosphate (day 168) conditions. In order to compare the different transcription profiles of the two MAGs, the analysis was conducted at the gene ortholog levels. Table 2 summarizes the number of differentially transcribed KO in the different comparisons. An orthologs was defined as up or low-transcribed if it was differentially transcribed in either A. delftensis or A. phosphatis. The results are summarized in Figure 4 for the feeding phase and in the Supplementary Figure 9 for the aerobic phase.

It can be noticed that the differences in the gene’s transcription between the Accumulibacter types were the same in both phosphate conditions, suggesting an inherent difference between the types. Compared to Accumulibacter phosphatis type II, Accumulibacter delftensis type I up-transcribed the genes from the phosphate-related pathway, the EMC, the VFA activation via the phosphate acetyltransferase (pta) and the acetate kinase (ackA), the sucABCD genes from the TCA cycle, and the glyoxylate shunt. Conversely, A. phosphatis type II up-transcribed genes from the MMC pathway, the fumarate reductase (frdABCD) and succinate dehydrogenase iron–sulfur subunit (sdhB) from the TCA cycle, and the VFA activation via the acetyl-coenzyme A synthetase (acs) but only in the high-phosphate condition.

To elucidate how the transcriptomes of GAO Propionivibrio and Accumulibacter – which can display GAM and PAM – differed, and whether Accumulibacter performing GAM relied on different pathways than the GAO Propionivibrio, differential expression analysis was conducted at the gene ortholog levels. Table 3 and Supplementary Figure 10 indicate that around 80% of the KO transcribed by Propionivibrio were also transcribed by Accumulibacter.

The analysis was done to compare the transcription profile of the GAO P. aalborgensis to A. regalis when the latest one was under high-phosphate condition, considered behaving as a classical PAO (day 103), or under low-phosphate condition, considered to behave as a GAO (day 182). In both cases, Propionivibrio transcription was taken from day 182 as the sequencing depth was not high enough under high-phosphate conditions at day 103. Feeding and aerobic phases were compared separately.

The number of differentially transcribed KOs was lower when comparing Propionivibrio aalborgensis to Accumulibacter regalis under GAM mode than compared to Accumulibacter under PAM mode (Table 3). Moreover, only few of these KOs, 47 and 55 in feeding and aerobic phases respectively, were different when comparing Propionivibrio to Accumulibacter-GAM (Supplementary Figures 11A,B). Interestingly, the number of differentially transcribed KOs determined when comparing two types of Accumulibacter in GAM mode was higher than the number obtained when comparing Accumulibacter-GAM to the GAO Propionivibrio (Supplementary Figures 11C,D). However, at least half of these KOs were shared between the two comparisons.

Some transcriptional differences were observed in the EBPR-related pathways between Propionivibrio and Accumulibacter, as shown in Figure 5 and Supplementary Figure 12. Propionivibrio exhibited a higher transcription of the reductive TCA cycle (frd and suc genes) associated with the complete transcription of the MMC pathway compared to Accumulibacter in either high- or low-phosphate conditions. Conversely, Accumulibacter up-transcribed genes from the phosphate pathway, which would be expected as Propionivibrio, although needs to incorporate phosphate for survival, do not rely on the polyphosphate reserve for energy production. An exception was observed for the pst transporter and pho regulon which were up-transcribed by Propionivibrio when compared to Accumulibacter under high-phosphate condition. However, this could be explained by the low transcription of the high-affinity transporter in Accumulibacter when the concentration of phosphate is high, as described in section 3.3.

This study focused specifically on Accumulibacter species dynamics and metabolisms under different phosphate conditions. In order to optimize the enrichment of different Accumulibacter species, two SBRs with different media compositions were operated. Acetate and propionate or acetate and glucose were used as carbon sources in experiments RA and RC, respectively. Both volatile fatty acids, especially propionate, associated with a low COD/PO₄-P ratio, were previously acknowledged to favor the enrichment of Accumulibacter (Weissbrodt et al., 2013). The glucose uptake by Accumulibacter remains unclear, although some studies have suggested that some species are capable of utilizing glucose (Ziliani et al., 2023; Xie et al., 2024a). In addition to the carbon sources as a selective agent, allylthiourea (ATU) was added in experiment RA to inhibit the nitrification and subsequently the denitrification. It is well recognized that the denitrification capability differs between Accumulibacter species (Skennerton et al., 2015; Saad et al., 2016). The absence of ATU could have allowed the selection of denitrifying Accumulibacter.

The difference in the operational conditions could explain the observation of different Accumulibacter species between the two reactors. The lowest diversity in species observed in RC compared to RA can be due to the longer duration of the experiment and/or the medium composition. However, it is impossible to disentangle the influence of these different elements in the selection process due to the absence of control.

Nonetheless, the results indicate the persistence of Accumulibacter type I under low-phosphate conditions. If previous reports suggest that both Accumulibacter type I and II can shift their metabolisms from PAM to GAM, in most of these studies, mainly Accumulibacter type II remained the most abundant under low-phosphate concentration (Acevedo et al., 2012; Welles et al., 2015, 2016). One possible explanation given by Welles et al. (2015) was that Accumulibacter type I demonstrates a lower competitiveness than type II due to a reduced acetate uptake rate under polyphosphate-depleted conditions (Welles et al., 2015). However, when a longer acclimation period to the phosphate modification was applied, as done in the present study and by Acevedo and coworkers, Accumulibacter type I became the most active. The authors suggested that the acclimation period and the pH, which was not regulated and varied from 7 to 9, could influence the distribution of the different Accumulibacter types to shift from PAM to GAM (Acevedo et al., 2017). The pH is a factor well recognized to influence the kinetics of the carbon uptake between the microorganisms (Zhang et al., 2007; Weissbrodt et al., 2013). Furthermore, the potential use of different pathways between the two types can explain the different adaptation capabilities of Accumulibacter types. Accumulibacter type I enhancing the low-affinity acetate activation, phosphate-related pathway and the EMC pathway for redox balancing, while Accumulibacter type II seemed to rely more on the MMC pathway. These results are described in more detail in the following sections.

In experiment RA, the GAO Propionivibrio became highly abundant under low-phosphate conditions. In previous studies, no GAO were detected in high abundance in high- or low-phosphate conditions. However, FISH probes targeting some other GAOs, such as Defluviicoccus and Candidatus Competibacter, were used to estimate their abundance, albeit specific probes targeting Propionivibrio were not used. Moreover, the use of the PAO-mix probes (PAO492, PAO651, PAO846) can be misleading, as these probes target Propionivibrio, as shown by (Albertsen et al., 2016). The enrichment of Propionivibrio in the current study could have been the effect of the soft conditioning cycle operated on the RA reactor, i.e., settling and withdrawal operated before the aerobic phase for two cycles repeated twice in two weeks, whereas no conditioning cycle was operated in RC. In their studies, Zhao et al. (2022, 2023) operated their reactors with these inverted phases over an extended period. In their case, they obtained an enrichment of the GAO Candidatus Contendobacter, which may have been favored over the other GAO by the longer period of operation and/or different operational conditions (e.g., acetate as sole carbon source, non-controlled pH).

Our results showed that not only the pH or the duration of the experiment, but also the carbon source, the presence or absence of ATU, and the flanking community members (i.e., presence of GAO), can play a role in Accumulibacter type distribution under low-phosphate conditions. These different operational parameters can possibly trigger specific metabolic activities and should not be neglected while interpreting the metatranscriptomics results.

The metabolisms of the different populations were compared under the different conditions and discussed below. A schematic of the transcription profiles of Accumulibacter under different phosphate conditions and Propionivibrio is proposed in Figure 6 and Supplementary Figure 13.